topic Re: How to get certain quantile when fitting a distribution on a dataset? in SAS Programming

How to get certain quantile when fitting a distribution on a dataset?

Al_senior — Sat, 24 Jul 2021 15:11:37 GMT

I am re-posting this question in hope of getting any answer.

I am fitting different distributions to a data set. I am using the following code:

proc capability data=pdlsf;
 specs usl=usl;
 var result;
 histogram / normal(indices) Lognormal(indices);
  by par1 par2;
 inset mean std cv n ppk="Ppk" normal(ESTPCTGTR) normal(ksdpval)/ cfill=white pos=ne;
run;
quit;

When running the code as expected I get a nice quantile table with predefined quantiles like

1%, 5%, 10%,...., 99%.

What I need is to get some certain quantiles that is not predefined. Let's say for some calculation I need to get 99.22% quantile estimated by lognormal.

Can I get that non-pre-defined quantile with proc capability? if yeas, how? If not, what proc step can do that for me?

A desired outcome could be something like the following table which is basically a quantile table generated by proc capability step with only difference that I defined 99.22% to be calculated in this table.

                   Quantiles for lognormal Distribution

quantile                  observed                   estimated
1                           12                       11.422
99.22                       21                       19.485

Re: How to get certain quantile when fitting a distribution on a dataset?

jimbarbour — Sat, 24 Jul 2021 18:44:45 GMT

I'm not familiar with Proc Capability, so this may be of no help at all, but here is a paper that discusses the best SAS procedures for generating custom percentiles: https://www.lexjansen.com/wuss/2014/114_Final_Paper_PDF.pdf

It might be that you would need to define/determine the distribution using Proc Capability and then the percentiles with another procedure. A bit of a pain to use multiple procedures, I know, but since no one else has responded, I thought I'd at least suggest supplemental procedures.

Jim

Re: How to get certain quantile when fitting a distribution on a dataset?

PGStats — Sat, 24 Jul 2021 21:31:32 GMT

Here is how to get the observed quantile which does not depend on the assumed distribution:

proc capability data=sashelp.heart;
var cholesterol;
output out=perc n=n median=med geomean=geom pctlpts=99.22 pctlpre=pChol PCTLNDEC=2;
run;

proc print data=perc label noobs; run;

Re: How to get certain quantile when fitting a distribution on a dataset?

Ksharp — Sun, 25 Jul 2021 11:36:39 GMT

proc univariate data=sashelp.heart CIPCTLDF noprint;
var cholesterol;
output out=perc n=n median=med pctlpts=99.22 pctlpre=pChol;
run;

proc print;run;